In this paper, a new neuro-musculoskeletal gait simulation platform is presented. This platform is developed to reproduce healthy or altered walking gaits. It is based on an original model of central pattern generator able to generate variable rhythmic signals for controlling biological human leg joints. Output signals of motoneurons are applied to excitation inputs of modelled muscles of the human lower limbs model. Eight central pattern generators control a musculoskeletal model made up of three joints per leg actuated by 44 Hill-type muscle models. Forward dynamics simulation in OpenSim show that it is possible to generate different stable walking gaits by changing parameters of controller. Further work is aimed on development of stable human standing by implementing reflexes.